Physical rehabilitation is the main treatment for motor disorders. Yet we are far from understanding the essential principles of rehabilitation training in adults, and know even less about children. Our long-term goals are to identify critical features of the motor learning process, understand how they change during development, and discover how to optimize them for rehabilitating children and adults with hemiparesis. Here we focus on optimizing motor learning during split-belt treadmill training, a technique that shows strong promise for producing dramatic improvements in walking patterns. We have shown that deficit exaggeration during split-belt training can correct walking deficits in children and adults with hemiparesis. Specifically, exaggerating the problem by making one leg walk faster than the other drives the nervous system to adapt, and this improves walking patterns when the person returns to normal conditions. We believe that split-belt deficit exaggeration could become an important clinical approach. However, there are two critical motor learning features requiring further study to guide clinical application. First, we need to optimize generalization from split-belt walking to walking under natural conditions. Split-belt adaptation generalizes only partially to natural over-ground walking. This is a common problem with any device or robot based training strategy, where training conditions do not match real-world contexts. Thus, we will study how manipulating the sensory context with virtual reality and changing attention to the movement influence generalization. Second, we need to optimize retention of adapted walking patterns, in order to produce long-term improvements in natural walking. In previous experiments, adaptation faded within 10 20 minutes. In the proposed experiments, we will study how attention and training schedule can be used to extend retention. In all three aims, we will assess effects of age (3-adult) and presence of hemiparesis from cerebral damage. Overall, we believe our results will show how deficit exaggeration during split-belt treadmill training could be used to produce long-term improvements in natural walking.